[DRBD-user] DRBD 8.4.1 fails to bring up stacked resource on debian 2.6.32-5-amd64

envisionrx ron.wells at envision-rx.com
Tue Jan 31 18:03:04 CET 2012

Note: "permalinks" may not be as permanent as we would like,
direct links of old sources may well be a few messages off.



Lars Ellenberg wrote:
> 
> On Mon, Jan 30, 2012 at 02:56:57PM -0600, Ronald Wells wrote:
> 
>>Only that 8.4.0 is much more seriously broken than 8.4.1 :-/
> 
> I was afraid of that.. I am planning to go back to 8.3 branch, but wanted
> to post this in hopes it would help fix a bug.
> 
> 
>>I suggest you retry and configure
>>	"no-disk-barriers; no-disk-flushes; no-md-flushes;"
>>for at least the stacked resource,
>>that may help working around this issue.
> 
> If doing that would help debug the prob I'm happy to, I'm assuming that
> would not be recommended for a production system?
> 
>> Jan 27 16:22:43 drbd kernel: [  191.128580] Call Trace:
> 
>>> Jan 27 16:22:43 drbd kernel: [  191.128602]  [<ffffffffa02882fc>] ?
>>> drbd_make_request+0x2b/0x14b [drbd]
>>> Jan 27 16:22:43 drbd kernel: [  191.128605]  [<ffffffff81180eeb>] ?
>>> bio_end_empty_barrier+0x0/0x24
> 
>>This does not compute.
> 
>>Why would bio_end_empty_barrier() call drbd_make_request?
>>Or, why would bio_end_empty_barrier() call anything at all?
>>That's an atomic bio completion function that set at most two bits,
>>and does one completion.
> 
>>It can not possibly wait for anything (but the spinlock of the
>>completion; which would not be reported as "soft lockup", but as a
>>spinlock deadlock).
> 
>>The call path actually should be
>> bm_rw -> blkdev_issue_flush -> submit_bio -> generic_make_request ->
>>drbd_make_request -> bio_endio(, -EOPNOTSUPP), which ends up in
>>bio_end_empty_barrier(), which, as indicated above, should be atomic.
> 
>>Configuring DRBD with no-md-flushes etc. as indicated above would
>>not even try to call blkdev_issue_flush, thus skipping this completely.
> 
> Good question.  Would it be helpful for me to try that and post the
> results, or not since it appears to be a bug?  Are there additional logs
> or anything else that would be helpful in trying to debug this?
> 
> 
> 

-- 
View this message in context: http://old.nabble.com/DRBD-8.4.1-fails-to-bring-up-stacked-resource-on-debian-2.6.32-5-amd64-tp33234353p33237674.html
Sent from the DRBD - User mailing list archive at Nabble.com.




More information about the drbd-user mailing list